AetherSDR v0.9.8 — Aetherial Audio Channel Strip RX + community-driven reliability sweep

Aetherial Audio Channel Strip RX + community-driven reliability sweep

A focused 24-hour sweep around the Aetherial Audio Channel Strip's RX
side and a long list of community-contributed fixes. The RX panel
gains DESS, full final-output / waveform stages, ADSP cluster bypass,
and parity with the TX layout. Hamlib NET-rigctl interoperability is
unblocked end-to-end — Not1MM, MacLoggerDX, fldigi, and any other
standard rigctld client can now drive AetherSDR's frequency, mode,
filter passband, RF power, PTT, and CW. The release also clears two
silent Windows footguns (UI Scale on Windows was completely broken;
spot clients crashed on disconnect cascade), and lands a CW-knob UX
upgrade plus a dialog chrome refit on the AetherDSP Settings window.

Big thanks to @chibondking (CJ Johnson — editable CW value fields,
cwDelay optimistic-update, UI Scale Windows + reset-to-100% fix, macOS
relaunch via open -n, PGXL OPERATE/STANDBY sync), @jensenpat
(spot client thread affinity Windows crash, hardware-PTT mic_selection
fix, filter width indicator, mode-correct widen/narrow shortcuts,
Not1MM rigctld interop), @rfoust (Robbie Foust — status bar
separator dot fix, audio shutdown null-guards), and @s53zo (MIDI
VFO jog-wheel stabilization). Also crediting @ct1drb for the
original orphaned rigctld pipe-separator fix that #2438 cherry-picked.

New features

Aetherial Audio Channel Strip — RX side (#2425)

• The Channel Strip now has a fully-built RX path alongside the TX
  path. RX side mirrors the TX layout: AGC-G (formerly AGC-T), EQ,
  De-Esser, Compressor, Tube/Saturator, PUDU, Final Output, and
  Waveform tiles, all wired through the inline RX DSP pipeline in
  AudioEngine.
• The CHAIN status row's ADSP tile is now a Stage-style toggle that
  bypasses the entire client-side NR cluster (NR2 / NR4 / MNR / DFNR
  / RN2 / BNR) with a single click; an in-memory snapshot restores
  the prior NR state on un-bypass.
• ChannelStripPresets schema gains an optional rx block so saved
  profiles capture both TX and RX state. Old preset files without
  the block leave RX state untouched (forward-compatible).
• New StripRxOutputPanel with peak/RMS meter, MUTE, BOOST. Strip-side
  Waveform panel becomes side-aware so the same widget renders the
  TX or RX scope tap based on a Side::Tx | Side::Rx flag.

Editable CW Delay / Speed / Sidetone / Pitch fields (#2429, chibondking)

• The four CW value labels in the PhoneCwApplet are now QLineEdit
  widgets. Click any value and type a number directly — SmartSDR
  parity. QIntValidator clamps to the valid range, editingFinished
  drives the slider so the existing slider→model command path fires
  unchanged. hasFocus() guards on the slider valueChanged path AND
  on syncCwFromModel keep mid-edit input from being clobbered by
  radio echoes.

Hamlib NET-rigctl / Not1MM interoperability (#2438, jensenpat, closes #2048 #2108)

• Pipe separator (|) — rigctld's wire convention. AetherSDR only
  recognised ;, so every pipe-prefixed command returned RPRT -4.
  Cherry-picks the orphaned PR #2051 fix from @ct1drb with full
  attribution.
• Multi-line bare b send_morse — Hamlib spec allows the morse text
  on the next line. Per-connection m_pendingMorseLine flag arms the
  next handleLine call to consume morse text verbatim. Required by
  Not1MM contest CW.
• RFPOWER get_level / set_level — was returning RPRT -11 (silent
  failure). Wires through to TransmitModel::setRfPower / rfPower()
  with proper 0.0–1.0 ratio ↔ 0–100 percent conversion. dump_state
  set_level mask was hardcoded to KEYSPD only — now correctly reflects
  the full RFPOWER + KEYSPD set.
• Per-mode filter passband placement — cmdSetMode was applying the
  USB convention to every non-LSB mode, so M CW 250 produced a
  155 Hz filter centered at 172 Hz instead of a 250 Hz filter centered
  on 0. Mirrors the canonical mode→filter-edge mapping from
  VfoWidget::applyFilterPreset.

AetherDSP Settings dialog refit + VFO DSP launchers

• AetherDSP Settings dialog gets the standard frameless 18 px gradient
  title bar (matching NetworkDiagnosticsDialog and AetherialAudioStrip)
  with grip glyph, min/max/close trio, drag-to-move, double-click to
  maximize, 8-axis resize.
• New setDialogMode(true) on AetherDspWidget bumps every inline
  font-size to 13 px to match the VFO DSP toggle row, applies a
  parallel 60×24 / 70×26 toggle-button geometry. Applet path leaves
  this off.
• Per-slice VFO DSP grid gains an ADSP button (opens AetherDSP
  Settings — same entry point as the Settings menu) and an
  AetherVoice button (toggles the Aetherial Audio Channel Strip).

Bug fixes

Spot client thread affinity Windows crash (#2420, jensenpat, fixes #1929 #2418)

• QTcpSocket / QUdpSocket / QWebSocket create an internal
  QSocketNotifier whose Win32 message-loop affinity binds to the
  construction thread, not the QObject's current thread. Constructing
  on main-thread then moveToThread() to SpotClients left the
  notifier bound to main-thread; disconnect-cascade socket events
  tripped a cross-thread sendEvent assert. Each client now defers
  socket creation to an initialize() slot dispatched on the worker
  thread via Qt::QueuedConnection so internal notifiers bind there.

Audio callbacks during shutdown (#2413, rfoust)

• Quit-time crash on macOS where WanConnection::~WanConnection
  triggered a TLS disconnect cascade that emitted
  radioTransmittingChanged(false) after m_audio was already null
  but before QObject::~QObject had auto-disconnected the connected
  lambdas. Surgical null guards on the three TX-state callbacks
  (moxChanged, txAudioGateChanged, radioTransmittingChanged)
  match the existing if (m_audio && ...) pattern used elsewhere in
  the file.

Hardware PTT into radio with mic_selection=PC (#2431, jensenpat, fixes #2373 #2200)

• The interlock gate was forcing setTransmitting(false) whenever
  m_txRequested was false — even when the radio reported our handle
  as TX owner. Hardware-keyed PTT (mic-line, ACC footswitch, RCA
  TXREQ) had no bypass. PC mic audio with hardware PTT had likely
  never worked through this branch. Parses the radio's interlock
  source= field (SW / MIC / ACC / RCA / TUNE per FlexLib v4.2.18
  ParsePTTSource) and adds MIC|ACC|RCA to the bypass list. SW
  source still requires m_txRequested so SSB optimistic-off
  responsiveness is preserved.

UI Scale not applied on Windows; reset to 100% silently ignored (#2432, chibondking)

• Two unrelated bugs in main.cpp's pre-Qt settings read. (1) The
  pre-Qt path was hardcoded to ~/.config/AetherSDR/ (Linux
  convention); on Windows AppSettings actually writes to
  %APPDATA%/AetherSDR/, so the file was never found and
  QT_SCALE_FACTOR never set. Adds a Q_OS_WIN branch using
  qEnvironmentVariable("APPDATA"). (2) The if (pct != 100) guard
  meant a child process inheriting QT_SCALE_FACTOR from a scaled
  parent kept using it after the user reset to 100%. Always-set fix
  writes 1.00 so the inherited value is overridden.

macOS relaunch via open -n on UI Scale restart (#2434, chibondking)

• applyUiScale() was calling QProcess::startDetached on
  applicationFilePath(), which on a .app bundle launches
  Foo.app/Contents/MacOS/Foo directly — bypassing Launch Services.
  The relaunched instance shows as a separate dock entry, loses
  proper activation policy, and on notarized/sandboxed builds can
  fail entirely. Walks up to the bundle root and launches via
  open -n <bundle> with --args pass-through.

MIDI jog-wheel VFO tuning stabilization (#2422, s53zo, fixes #2421)

• Three independent failure modes on touch-sensitive jog wheels (e.g.
  Hercules DJControl Starlight). MIDI Learn could bind the touch
  sensor instead of the movement CC. Saved bindings without the
  relative flag interpreted unit-pulse 1/127 as absolute endpoints,
  turning each detent into ~±63 steps. Tuning targets read
  SliceModel::frequency() which lags behind in-flight tune commands,
  making rapid spins jump backward. Fixed in three layered passes
  with a 250 ms idle-timer-cleared in-flight tune target.

VFO filter width indicator showing 0.1 kHz too low (#2435, jensenpat, fixes #2197)

• updateFilterLabel() computed raw filterHigh - filterLow, which
  for SSB/digital modes is ~100 Hz less than the named preset width
  due to the intentional low-cut. Bug had been reported three times
  (#794, #1225, this issue) — the structural fix promotes
  RxApplet::formatFilterWidth to public static and has VfoWidget
  call it directly. Single source of truth prevents the next swap.

Mode-correct filter widen/narrow shortcuts (#2436, jensenpat, fixes #2208)

• The filter_widen / filter_narrow shortcuts (also bound to MIDI
  global.filterWiden / global.filterNarrow) only adjusted the
  upper edge by ±100 Hz. On LSB / CWL / DIGL / RTTY the upper edge is
  fixed near 0, so "widen" actually collapsed the passband. New
  RxApplet::stepFilterWidth(direction) walks the active mode's
  preset list and routes through applyFilterPreset so all modes get
  mode-correct edge geometry.

PGXL OPERATE/STANDBY button stuck on OPERATE (#2437, chibondking)

• AmpApplet button was only updated via the direct PGXL TCP path; if
  that lagged, the button stayed on the old label. The click handler
  reads m_operateBtn->text() to decide toggle direction, so a
  stuck OPERATE button emitted operate=0 on click — sending another
  standby command instead of returning to OPERATE. Wires
  RadioModel::ampStateChanged (authoritative) to also call
  AmpApplet::setState.

Status bar separator dots for unsupported optional sections (#2412, rfoust)

• TGXL and PGXL separator dots stayed visible even when the indicators
  were hidden, leaving stray · · clusters in the bottom status bar
  on radios without those accessories. Optional separators now toggle
  in lockstep with their matching indicators.

cwDelay optimistic-update prevents Delay slider reset (#2428, chibondking)

• setCwDelay() was the lone outlier in the CW setter family —
  didn't cache ms into m_cwDelay before emitting the command.
  Subsequent phoneStateChanged emissions read stale m_cwDelay
  and snapped the Delay slider back, silently re-enabling QSK on
  amplifiers that can't tolerate it. Now matches the
  setCwBreakIn / setCwIambicMode optimistic-update pattern.

DSP-level slider missing on launch

• The leveled DSP toggles (NR / NB / ANF / NRL / NRS / NRF / ANFL)
  push onto a target stack so the shared 4th-row slider re-targets
  the most recently activated DSP. State changes arriving from radio
  status (profile load, reconnect) went through a QSignalBlocker'd
  path that suppressed the toggled() signal — so on launch the
  slider was hidden until the user manually toggled the DSP off and
  back on. New connectLeveledDsp helper mirrors the user-click
  stack push/pop on every state change.

Channel Strip initial height on sub-4K displays (#2440)

• The Aetherial Audio Channel Strip opened at its natural 1620 px
  height on every display. On 1080p the bottom edge landed offscreen
  with the resize grip unreachable. Queries
  QGuiApplication::primaryScreen()->availableGeometry().height()
  at construction; caps initial height to 960 px on anything below
  ~4K (2000 px threshold catches 1080p / 1440p / 1600p). 4K and 5K
  still get the natural 1620.

Infrastructure

CI runtime + reliability sweep

• Skip CodeQL on doc-only PRs (#2397).
• Cache Windows third-party deps — Opus, FFTW, hidapi, DeepFilterNet
  (#2398) — drops cold Windows builds by ~3 min.
• Mirror Opus download to GitHub Releases with retry/fallback (#2399)
  — replaces the unreliable Xiph mirror that occasionally 503s.
• Scope CodeQL to first-party sources (#2400) — drops third_party/
  noise from the analysis.
• Speed up macOS Intel build with pinned Qt + ccache (#2402).
• Bump Qt to 6.8.3 LTS across all release workflows (#2404).